基于改进BP神经网络的油井流入动态研究

Study on the inflow performance of oil wells based on improved BP neural network

针对传统流入动态研究方法误差较大等问题,提出利用改进的BP神经网络进行油井流入动态研究。通过采用LM算法和贝叶斯正则化算法,改进了常规BP神经网络收敛速度慢、泛化能力差等缺点,并探讨了通过优化网络结构来提高网络泛化能力的方法。实例计算结果表明,采用LM算法和贝叶斯正则化算法的改进BP神经网络用于油井流入动态研究是可行的,且优于传统的流入动态研究方法,具有精度高、收敛速度快、泛化能力强等特点。

To reduce the errors in traditional research methods for inflow performance relationship （IPR） , a new approach was proposed to predict IPR of oil wells by using the improved BP neural network. In this new method, the LM algorithm and the Bayesian regularization algorithm were adopted to solve some problems in general BP algorithm, such as slow convergence speed and poor generalization performance. And the generalization performance was improved by optimizing network structure. The Example computed result showed that the new method has high precision, fast convergence speed and strong generalization performance. It is superior to traditional methods.